Deep learning algorithm is more accurate in predicting which patients with knee osteoarthritis might need a total knee replacement.
Total knee replacement is a significant surgery that potentially anyone with knee osteoarthritis might face. With the advent of a new artificial intelligence (AI) model, providers will have a better idea about which patients are likely to need the procedure.
With more than 14 million Americans living with knee osteoarthritis (OA), and more than 50 percent ultimately needing a total knee replacement, being able to more accurately predict which patients face a trip to the operating room can help providers determine the best treatment strategies.
To help clinicians with this assessment a research team, including imaging and data experts, from New York University designed and tested an AI model with more than 700 participants. Their results, published in the June 23 Radiology,indicate that, not only is their model better than current prediction methods, but it could also be used to create disease-modifying therapies designed to prevent (or at least postpone) knee surgery.
“We developed a model to predict OA progression outcomes directly from baseline radiographs,” wrote the team, led by Cem Deniz, Ph.D., assistant professor of radiology at New York University Langone Health. “However, we developed a model to predict OA progression outcomes directly from baseline radiographs with an additional [grading system] prediction task.”
Specifically, the team retrospectively analyzed medical information from 728 individuals – 324 patients with knee osteoarthritis and 324 without the condition who were enrolled in the multi-center, longitudinal Osteoarthritis Initiative. The data evaluated included X-rays and MRI images.
Based on their results – an area under the receiver operating characteristic curve score of 0.87 – the team determined the deep learning model was effective. In fact, it outperformed other system currently used to determine which patients are best suited for total knee replacement, including the Kellgren Lawrence grade and the OA Research Society International atlas.
In an accompanying editorial, Michael L. Richardson, M.D., professor of radiology and orthopedic surgery at the University of Washington in Seattle, noted that while this deep learning model is not the first AI model to predict the likelihood of surgery among patients, its performance is among the highest.
“[This] deep learning model…goes a step further, cuts out the middleman (KL classification) and predicts [total knee replacement] need directly from the patients’ knee radiographs,” he wrote.
Further evaluation is needed with external datasets before this AI model can be used in an actual patient-care setting, but its development could be a huge benefit for patients living with knee OA.
“The availability of an automated system that could accurately predict [total knee replacement] need has important and practical ramifications,” he wrote. “There is, therefore, a great necessity to begin disease-modifying therapies as soon as possible to prevent or delay the need for a [total knee replacement.]”
Stay at the forefront of radiology with the Diagnostic Imaging newsletter, delivering the latest news, clinical insights, and imaging advancements for today’s radiologists.
The Reading Room Podcast: A Closer Look at Remote MRI Safety, Part 2
July 25th 2025In the second of a multi-part podcast episode, Emanuel Kanal, M.D. and Tobias Gilk, MRSO, MRSE, share their perspectives on remote MRI safety protocols for ensuring screening accuracy and adherence to conditional implant guidelines as well as a rapid and effective response to adverse events.
The Reading Room Podcast: Current and Emerging Insights on Abbreviated Breast MRI, Part 2
July 23rd 2025In the second part of a multi-part podcast episode, Stamatia Destounis, MD, Emily Conant, MD and Habib Rahbar, MD, discuss key sequences for abbreviated breast MRI and how it stacks up to other breast cancer screening modalities.
FDA Clears New Radiography/Fluoroscopy Systems from Siemens Healthineers
July 21st 2025Offering ergonomic controls and AI-enabled features, the remote-controlled Luminos Q.namix R system and the Luminos Q.namix T platform with tableside control reportedly facilitate workflow efficiencies for complex radiography and fluoroscopy examinations.